Manufacturing Fuels Supercomputing Growth
Big rigs have become a staple of the American highway, but lately they've become an uncomfortable metaphor for our country as a whole. They're big, they're wasteful, and harmful to the environment.
This problem led Mike Henderson, president of SmartTruck to drastically rethink the way his trucks were made. “Trucks,” he explained, “are fairly unaerodynamic devices. Half their fuel goes to defeating aerodynamic drag.”
And because of air quality legislation in California, Henderson is looking for a solution for any of his trucks passing through the state. The law requires that any truck must feature low rolling resistance tires and approved aerodynamic devices that will help to cut down on the rigs' gas guzzling and emissions.
But developing addons that meet the EPA SmartWay regulations require extensive simulations on a multitude of components and assemblies—the sheer complexity of which a regular workstation can't handle.
“Almost every piece of a truck produces wake and they mix with other wakes,” said Henderson in an interview with NPR. “It would take over a week to do a single calculation” to simulate a part's interaction with the rest of the truck.
So Henderson looked to borrow a supercomputer that solve these problems for him. Specifically, he went to the Department of Energy's Oak Ridge National Laboratory, which houses Titan, “the most powerful computer in the world for open science,” according to Suzy Tichenor.
We've already seen trucking companies take advantage of high performance computing to meet the DOE's goals for its Super Truck Program, which aims to improve big rig fuel economy by 50 percent. Even SmartTruck has gone to Oak Ridge in the past to use its Jaguar super.
Nonetheless, Titan is a major step up, offering roughly 27 petaflops, or 27 quadrillion calculations (or operations) per second to Jaguar's 1.7 petaflops. And just for context, a quadrillion is a million billion.
“It's a big boy,” said Tichenor.
And businesses of any size—even small businesses like Henderson's—can apply to tap its power. And if the companies make their findings public, the service is free.
On top of being able to test a prototype truck without first manufacturing it, HPC simulations have the added benefit of allowing engineers to clearly view all aspects of airflow that would have been invisible in a wind tunnel.
“It probably got us to market a year faster than we would have had we not used it,” Henderson said.
Two years ago, when SmartTruck went to Oak Ridge for its Jaguar super, it was looking to speed up the development of its UnderTray System, which funnels air away from the wheels to reduce drag. Using HPC clusters already at the company's disposal would have meant modeling one part of the system at a time, which would take four days for each component. But with Jaguar, the simulation time went from weeks down to hours, demonstrating what supers of any size have to offer.
And other companies with complex engineering problems have taken note. Over the past year alone, Tichenor said that the number of Titan applicants has increased 20 percent.
“We've definitely seen a growth in the number of companies that are applying because modeling and simulation allows you to accelerate the research and development process dramatically,” she said.
And the HPC adoption doesn't stop there. According to Earl Joseph with international market research company IDC, “the market has been growing at a rate we've never seen before.” Despite pricetags upwards of $10 million for a mid-sized super, Joseph said that even the recession hasn't made a dent in their sales.
“Supercomputers grew 65 percent in 2009 alone, last year sales grew another 29 percent. You're not seeing those growths in any other sector right now,” said Joseph.
The next big innovation in supercomputing that could be coming to manufacturers' doorsteps is “exascale computing,” where the machines are a thousand times faster than Titan. But oddly enough, the greatest hurdle keeping these from reaching businesses won't be their pricetags, per se.
“The biggest problem right now is the electrical bill,” Joseph said, referring to the $30 million to $50 million bill that an exascale machine would rack up annually. “To build that big of a computer you almost need a nuclear power plant right next to it.”
But once someone inevitably finds away around it, he says the race for an even greater super will press on.